Square MCP Server

The Square MCP Server connects AI assistants such as Claude, Cursor, Goose, or Windsurf to the Square API. Anyone who runs a shop, a café, or a service business on Square can ask the assistant questions like "Which orders came in today?", "Create a new customer", or "Which catalog items have no price?" without switching to the dashboard or writing their own code against the API. The server is operated by Block, the company behind Square. According to the documentation it is currently in beta.

What is the Square MCP Server?

MCP, the Model Context Protocol, is an open protocol through which AI assistants call tools of external services. The Square server takes an unusually lean approach: instead of offering a separate tool for every function, it provides three tools according to its README. With get_service_info the assistant learns which methods a Square service offers, with get_type_info it sees the parameters and data structures of a method, and with make_api_request it performs the actual call. As a result the server covers, according to the repository, more than forty services, including payments, customers, orders, inventory, appointment bookings, and loyalty programs. The repository is generated from Square's OpenAPI specification, so changes to its scope arrive through the generator.

What is it particularly useful for?

The greatest benefit lies in everyday administration and in evaluations that take several clicks in the dashboard. The assistant can search and complete customer records, summarize the orders of a period, check catalog items, or look up bookings. Developers benefit as well because the assistant explores the API structure itself: it can sketch integrations or formulate test calls without first working through the reference. Compared with a generic HTTP tool, the server knows the Square types and performs requests with correct authentication.

Remote server or local installation

Square recommends the hosted remote server at https://mcp.squareup.com/mcp. It signs you in through OAuth with your Square account, so you do not manage tokens by hand, and you approve only the permissions that are needed. According to the documentation the remote server works exclusively with production data. Anyone who wants to test safely first starts the local server with npx square-mcp-server start. It requires Node.js and an access token, but offers the sandbox mode (SANDBOX=true), in which no real customer or payment data is affected. With DISALLOW_WRITES=true the local server can additionally be restricted to read-only access.

Setup

For Claude Desktop, according to the documentation, you register the server through mcp-remote: the command npx with the arguments mcp-remote and the server URL. Cursor, Windsurf, Goose, and claude.ai are also named as supported clients. Square maintains an allowlist of MCP clients to prevent malicious client registrations; a client that is not listed has to be approved by Square before sign-in works.

Security and limits

The assistant acts with the permissions you grant at sign-in and can also trigger write calls through make_api_request, such as changing customers or creating orders. Square explicitly advises getting familiar with the server first and testing requests against the sandbox before using it in production. Confirm write actions deliberately and grant only the permissions the use case needs. Customer data the assistant retrieves enters the AI client context and therefore reaches the model provider; this has to be assessed under data protection law for your customers.

Costs

For the server itself the documentation names no separate fee. A Square account is required; for payments and additional services the regular Square terms apply, which you can find on Square's official pricing page.

Who is the Square MCP Server for?

It is worthwhile for merchants and service providers who already work with Square and want to handle routine tasks or evaluations by voice or text, and for developers building Square integrations. Anyone who does not use Square gains nothing from it. Keep the beta status in mind: scope and behavior may still change.

A Square account and an MCP client that supports remote servers (such as Claude Desktop, Cursor, or Windsurf). For the local variant additionally Node.js and an access token.

With the Square MCP Server, an AI assistant accesses customers, orders, catalog, and other Square data. This guide leads from choosing the variant to the first queries.

Choose a variant

Square offers a hosted remote server and a local installation. According to the documentation the remote server at https://mcp.squareup.com/mcp works only with production data and signs you in via OAuth. The local variant runs on Node.js, needs an access token, and offers a sandbox mode. For first experiments the sandbox is the safer choice.

Step 1: Test locally in the sandbox

Create a sandbox access token in your Square developer account. Start the server with npx square-mcp-server start and set ACCESS_TOKEN and SANDBOX=true. With DISALLOW_WRITES=true you allow read access only, which is enough for getting to know it.

Step 2: Connect the remote server

In the Claude Desktop configuration add a server with the command npx and the arguments mcp-remote and the URL https://mcp.squareup.com/mcp. On first connection the Square sign-in opens; approve only the permissions you really need.

Mind the allowlist

Square maintains an allowlist of MCP clients. According to the documentation, Claude, Cursor, Goose, and Windsurf are among those listed. Another client has to be requested from Square before sign-in succeeds.

Step 3: First queries

Start read-only: "List my ten latest orders." or "Which catalog items have no price?" The assistant calls get_service_info and get_type_info to find the matching method and then runs make_api_request.

Confirm write actions deliberately

Changes are also possible through make_api_request, such as new customers or orders. Square recommends testing requests in the sandbox first. Review every write action before you confirm it, and remember that retrieved customer data enters the AI client context.

npx mcp-remote https://mcp.squareup.com/mcp

Remote: OAuth sign-in with the Square account (recommended), only for allowlisted MCP clients. Local: access token in the ACCESS_TOKEN environment variable.

Read and write access to Square services within the permissions granted at sign-in, such as customers, orders, catalog, inventory, bookings, and payments. Locally, DISALLOW_WRITES=true restricts it to read access.

According to the documentation the remote server is hosted by Block. Retrieved customer, order, and payment data enter the AI client context and therefore reach the model provider.

Unintended write calls on production data, disclosure of personal customer data to the model provider, and misuse of access tokens in local operation.